A new Bayesian approach to nonnegative matrix factorization: Uniqueness and model order selection

摘要: Abstract NMF is a blind source separation technique decomposing multivariate non-negative data sets into meaningful basis components and weights. There are still open problems to be solved: uniqueness model order selection as well developing efficient algorithms for large scale problems. Addressing issues, we propose Bayesian optimality criterion (BOC) solutions which can derived in the absence of prior knowledge. Furthermore, present new Variational Bayes algorithm VBNMF straight forward generalization canonical Lee–Seung method Euclidean problem demonstrate its ability automatically detect actual number data.